SmolLM3-3B-SFT-ES

SmolLM3-3B-SFT-ES

This model is the result of the first stage of the ReasonXL two-stage reasoning adaptation pipeline applied to HuggingFaceTB/SmolLM3-3B.

Stage 1 — SFT

The model is supervised fine-tuned to shift its reasoning language from English to Spanish, using reasoning traces from toroe/ReasonXL-SFT.

The objective is to enable the model to perform its reasoning in the target language while preserving its reasoning capabilities.

Stage 2 — RL

The corresponding GRPO model is:

DGurgurov/SmolLM3-3B-SFT-GRPO-ES

The second stage applies RL (Dr. GRPO) to recover reasoning quality lost during SFT while preserving target-language compliance, using a composite reward over verifiable math problems.

Model Details

• Base model: HuggingFaceTB/SmolLM3-3B
• Target reasoning language: Spanish
• SFT dataset: toroe/ReasonXL-SFT
• Training stage: SFT
• Corresponding GRPO model: DGurgurov/SmolLM3-3B-SFT-GRPO-ES

Full training details, evaluation results, and methodology will follow soon.

Citation

If you use this model, please cite:

@misc{gurgurov2026reasonxlshiftingllmreasoning,
      title={ReasonXL: Shifting LLM Reasoning Language Without Sacrificing Performance},
      author={Daniil Gurgurov and Tom Röhr and Sebastian von Rohrscheidt and Josef van Genabith and Alexander Löser and Simon Ostermann},
      year={2026},
      eprint={2604.12378},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2604.12378},
}